Comparison of the performance of SVM and ANN to estimate water equivalent of snow height in East Azerbaijan

Snow and snowmelt are important factors controlling flow regimes in mountainous basins and provide most of the water. As a result, snow hydrology is of great value in mountainous areas. Estimation, simulation and prediction of flow resulting from rain and snowmelt has applications in provision of drinking and irrigation water, water control in rivers, flood control and flood warning systems, and estimations of flood damage in a basin. In this study, flow resulting from snowmelt was simulated using the MLP neural network and support vector machine (SVM) models for the Baladarreh Kandovan, Marand ski resortand, and Sandoghlu ski resort stations in the East Azerbaijan Province in the 2006-2013 period. The coefficient of correlation, root-mean-square error (RMSE), and mean absolute error (MAE) were used to evaluate accuracy. The results indicate that the neural network model had better accuracy compared to the SVM model. Among the different arrangements of neural networks, the 3-6-1 arrangement with snow density, length of snow sample, and snow depth had the highest accuracy.